Julia Muschalik's Threatening in English is a welcome contribution to recent research on threatening communications and a good addition to John Benjamins Pragmatics & Beyond New Series, which publishes interdisciplinary work within the language sciences. Muschalik's aim is ‘to report on [empirical research on the forms and functions of threatening speech acts] and to provide more insights into the nature of threatening gained by investigating a corpus of authentic American English threat utterances’ (p. 1). The organization of her work is easy to follow as it is presented in a traditional social science format (i.e. introduction, literature review, data and research design, analysis part 1 ‘form’, analysis part 2 ‘function’, and conclusion with suggestions for future research). The chapters are briefly summarized below and then followed by a discussion of a few notable limitations.

In chapter 1, Muschalik presents the problem of researching speech acts. Traditionally, speech acts have been defined by function and not form (Jucker Reference Jucker2009); therefore, they are difficult to identify and quantify in corpus data. She draws on previous literature in the field that has identified ways in which threatening language is ideologically constructed in order to test the forms and functions of threatening, ultimately to ask ‘How do threateners threaten?’ Her aim is twofold: (i) to identify forms, derive functional patterns from such forms and investigate if those forms can predict functions, and (ii) to apply a mixed method approach to advance ways in which pragmatic meaning can be investigated both quantitatively and qualitatively.

In chapter 2, Muschalik compiles features from previous studies in order to test their presence (or absence) in her corpus, which is similar to the first part of Gales' (Reference Gales2010) study that compiled untested assumptions about threatening language from three communities of practice and tested the frequency of those features in an authentic threatening language corpus. Here, Muschalik identifies six general categories of language that have been impressionistically associated with threats in previous literature – conditionality, futurity, violent verbs, pronouns, profanity and mention of weapons. And, since there is power in the act of uttering a threat (Gales Reference Gales2010) – i.e. power ‘in discourse’ (e.g. Culpeper Reference Culpeper2011; Limberg Reference Limberg2009) – Muschalik examines the functions of threats in relation to interlocutor power and distance. For example, threats have been found to be a strategy to manipulate or create conflict between interlocutors (Limberg Reference Limberg2009) or to convey impoliteness (Culpeper Reference Culpeper2011), both of which are inherently rooted in social hierarchies or institutional power – i.e. power ‘behind discourse’ (e.g. Culpeper Reference Culpeper2011; Limberg Reference Limberg2009). She hypothesizes that more powerful speakers (i.e. those with more social power such as judges, politicians and teachers versus those with less social power such as spouses, siblings and friends) will threaten with more direct (or pragmatically explicit) forms, a discussion that is revisited in her conclusion.

In chapter 3, Muschalik outlines her data and research design. In order to investigate the forms and functions of threats, she created the Corpus of Judicial Opinions (CoJO), which is comprised of 301 threats from 266 US Federal and State Judicial Opinions spanning the years 1918–2013. Of the 301 threats, 196 were spoken and 105 were written, with threats ranging in length from 2 to 496 words. While each threat is provided in the Appendix, the total word count of the corpus is not stated.

Regarding her data, Muschalik states the CoJO is authentic, in that it contains non-elicited data; is representative, as it contains threats from a range of regions, time periods and interlocutor goals; is pragmatically annotated, in that the threats were initially declared a threat by the recipient and cross-validated by a legal entity; and is contextually rich, in that judicial opinions contain detailed information about threatener target and sender.

The second part of chapter 3 presents the research design, which is described as a bi-directional analysis located in between formal and actional levels (i.e. between function-to-form and form-to-function levels) (Schneider & Barron Reference Schneider, Barron, Schneider and Barron2008). Muschalik states that her study will additionally test ‘the usefulness of different analytical tools for pragmatic analysis' and perform ‘a simplified descriptive analysis of linguistic forms’ using ‘a basic and reproducible coding’ (p. 48). While these methods are not described here, they are explained in the subsequent analysis chapters.

Chapters 4 and 5 provide analyses of the form of threats and the function of threats, respectively. In chapter 4, the questions addressed are summarized as follows (p. 52):

• • Are there typical features found in threats and do they correspond to what has been expected in the literature?

• • How frequent are the features and how are they distributed? Are any of the features pervasive in threats?

• • How do the features relate to each other? Are patterns of frequently co-occurring features meaningful?

In total, Muschalik coded for 12 variables: conditionality, futurity, type of verb, agent, patient, speaker orientation, hearer orientation, implicit hearer orientation, 1PP, 2PP, taboo language and mention of weapons. Italicized variables in this list were coded as binary (i.e. they were present or not); non-italicized variables were coded categorically (e.g. type of verbs: violent, non-violent, or ambiguous). The data were analyzed using a range of statistical measures in R and both significant (p < .05) and non-significant findings are discussed by each variable individually and, finally, as they co-occur with each other.

Muschalik's conclusions about the forms of threats both support and challenge previous findings on threatening language. For example, similar to other studies (e.g. Gales Reference Gales2010), taboo language was not found to be pervasive. But when it did occur, Muschalik highlighted an under-explored use of such language – to express emotion (in addition to the previously identified function of instilling fear). However, contrary to previous research, an analysis of verb types along with markers of futurity challenges earlier findings that categorize the majority of threats as veiled or indirect (Gales Reference Gales2010; Yamanaka Reference Yamanaka1995) and supports the idea that threats are pragmatically more explicit (i.e. direct) in some functional contexts.

Chapter 5 turns from the forms of threats to the function of threats. Muschalik's questions are summarized as follows (p. 130):

• • Can existing notions of threatening functions be confirmed by the data?

• • Can the contexts in which speakers threaten be categorized?

• • Can the form of a threat predict its function?

Drawing on Aijmer's (Reference Aijmer1996) work on the speech act of apologies, where apologies were coded as anticipatory (e.g. serving the functions of softening or disarming) or retrospective (e.g. serving the functions of supporting or self-demeaning), Muschalik outlines two macro-functions based on whether a threat is uttered pre-event or post-event and refers to them as threats of ‘manipulation’ and ‘retaliation’, respectively. Manipulative threats are uttered prior to an event in the hope of influencing the target's actions, which then affect the threatener's actions; retaliative threats are uttered after an event where the threat is a reaction to the target's actions, which then no longer influence the threatener's own actions.

Out of the 301 threats in her corpus, 139 were coded as manipulative and 162 as retaliative. She then performed two analyses: (i) each of the 12 forms from chapter 4 were quantified within the two functions, and (ii) two types of classification tree models were used to determine if any of the forms could be used as a predictor of the two functions.

While most findings were not statistically significant, they demonstrated a ‘slight disposition’ for manipulative threats to contain more conditional language, more futurate expressions (e.g. will) or no futurity, and more non-violent verbs. Additionally, they appeared to be more hearer-oriented with fewer first-person pronouns (p. 177).

In comparison, retaliative threats rarely contained conditional language, but contained more futurate expressions (e.g. be going to) and present futurate, and more violent verbs (e.g. kill or those resulting in death). These threats were also more speaker- or mixed-oriented and contained more taboo language (p. 177).

In terms of predictive strength, while conditionality was not found to be ‘a pervasive feature of threatening language in terms of its frequency’, it was found to be ‘a salient feature in terms of its functional association’ (p. 173) with a measure almost five times as high as the next category – verb type – in terms of predictive power (p. 17). The remaining features did not provide much, if any, predictive strength for differentiating the two functions.

Finally, and most directly related to the pragmatic question of ‘how threateners threaten’, Muschalik's analysis revealed that while manipulative threats are not very explicit, retaliative threats follow a more consistent pattern. Specifically:

Muschalik suggests that threateners' selections of such forms within the two functions may ultimately be related to speaker goals of indirectness in conflictive communication, which is discussed in the final chapter.

Chapter 6 concludes the research by refocusing on the pragmatic aspect of threatening as a speech act. Muschalik draws on the ‘strategic speaker’ model (e.g. Pinker et al. Reference Pinker, Nowak and Lee2008) to explore how her findings relate to explicit versus implicit ways of threatening. According to Pinker et al.’s (Reference Pinker, Nowak and Lee2008) research on bribery, ‘a strategic speaker makes numerous empirical predictions about how people interpret the intent and perceive the relationship between a speaker and a hearer depending on the level of indirectness in the speaker's words’ (Pinker et al. Reference Pinker, Nowak and Lee2008: 837 as cited in Muschalik, p. 188). In the traditional model, wherein both cooperative and antagonistic speakers are involved, indirectness is generally preferred. However, Muschalik's research demonstrates that the strategic threatener does not fit this traditional model. Instead, within threatening contexts, while manipulative threats were typically more indirect, retaliative threats were generally more explicit. Thus, more pragmatic research is called for on the functional use of speech acts in conflictive contexts.

Additional suggestions for future research include, for example, examining ‘less drastic forms of threatening’ to test the generalizability of the current results (p. 193), written forms to see if they are less pragmatically explicit, situations that presume a more hearer-oriented approach to threats, additional language varieties beyond American English, and other conflictive speech acts that may shed more light on impoliteness phenomena.

As an author whose own work is primarily focused on threatening language, I welcome and encourage original studies such as Muschalik's, which can advance our understanding of the speech act of threatening from a variety of perspectives. Much more work needs to be performed on threatening communications, and her exploration of form, function and pragmatic expression is a valuable contribution to the field. That said, I would like to comment on three important aspects of her study that must be considered when comparing and interpreting her results.

First, when comparing findings in research studies on threatening language, it is important to be aware of the research design in each case as differences may affect how the results are interpreted. The unit of observation in Muschalik's study is based on text, as opposed to author. That is, each text represents one unit of observation, so authors who have provided multiple threats have multiple units of observation, while others have only one (as documented in her Appendix). This may overrepresent certain stylistic features used by individual authors.Footnote ¹ Additionally, Muschalik's codings of each feature were based on binary or categorical decisions as opposed to normed frequency counts of the features within the units of observation. That is, her codings presented each variable as present or not present or as representing x, y or z category, as opposed to the frequency approach that examines the extent to which each feature is present (or not) within and across the units of observation. Both methods can provide interesting results, but when comparing studies with differing approaches, care needs to be taken when interpreting the conclusions.

Additionally, these choices in research design require us to revisit two of the research questions posed in chapter 4. Specifically: ‘How frequent are the features and how are they distributed? Is any feature so frequent as to reach the status of a pervasive conventional feature of threatening language?’ (p. 52). In using binary or categorical coding schemes, the question of the extent to which each feature occurs and its subsequent distribution by author cannot be adequately answered.

Second, corpus studies over the past two decades have repeatedly demonstrated the numerous differences between spoken and written language (e.g. Biber et al. Reference Biber, Johansson, Leech, Conrad and Finegan1999). While Muschalik addresses literature on both spoken and written language in her review, with approximately 200 spoken and 100 written threats in her corpus, she misses the opportunity to investigate – or even call attention to – possible differences or similarities between the two registers within her data and how they may affect both forms and functions. As she briefly notes in her ‘to-do list for future research’ in the conclusion, the investigation of spoken versus written registers of threatening communications is still an area in need of further exploration (p. 193), especially as previous work has commented on trends that demonstrate that some features of written threats are more similar to patterns found frequently in spoken non-threatening language (e.g. modal verbs in Gales Reference Gales2010; Nini Reference Nini2017).

Finally, and most critically, when examining the forms (and subsequent functions) of speech acts, it is essential that the data being examined are verbatim records of the language originally used to perform the speech act. Scholars studying forensic data will readily acknowledge that obtaining access to sensitive data can be extremely difficult. In this regard, Muschalik has been quite innovative in finding publicly accessible data that is pragmatically verified by two parties as threatening. However, as she herself briefly points out (pp. 45–6 and fn 31), there are many problems with relying on data that has been reported via personal recollection, retold by a second-hand witness, or transcribed by an uninvolved third party of the court.

According to the court opinions from which Muschalik gathered her data, of the 301 threats, 105 were written and 196 were spoken. That is, nearly two-thirds of the data were from the spoken register. Of those, only 25 were recorded on answering machines or other recording devices and transcribed by a member of the court. The remaining threats were claimed to have been reported ‘word for word’ by a witness as part of the evidence (p. 44). Witnesses included the threatener, target, bystanders and overhearers (p. 45).

Solan & Tiersma (Reference Solan and Tiersma2005) provide a good review of the literature on our abilities to recall language, especially in criminal and legal contexts. They summarize: ‘Our recall of the actual words that we read or hear is very poor’ and it is frequently influenced by the hearer's worldview and by inferences one draws from the context of the conversation. This is especially an issue ‘when someone is accused of committing a language crime, such as making threats’ (Solan & Tiersma Reference Solan and Tiersma2005: 99). In general, the studies they reviewed found that we are able to remember the gist, or meaning, of what is said, as opposed to the exact words that are uttered even a few moments after the utterance has been made. And when utterances are longer, even less is remembered.

Thus, while Muschalik states that spoken threats ‘were only extracted where they could be identified as complete and verbatim quotes’ (p. 43), given the fact that only 25 of the 196 spoken threats (13 percent) were recorded (and later transcribed by a court reporter, which also offers room for potential error), it is unlikely that the exact words – those used for the basis of this study – were actually captured verbatim in 87 percent of her spoken data. While her results may provide some very interesting trends in the forms and functions of threatening communications, further follow-up research needs to be performed in order to validate these findings.

In conclusion, Threatening in English by Julia Muschalik provides a well-organized, thoughtful analysis of the forms, functions and pragmatic contexts of the impolite speech act of threatening. Muschalik's study gathers features from previous ideologies about threatening language and tests them in her corpus of 301 spoken and written threats, which adds to the growing body of literature on threatening communications. While there are some notable limitations that need to be considered when interpreting her results, the trends she identifies in the forms and functions used by threateners and the pragmatic contexts in which they occur ultimately contribute to our further understanding of ‘how threateners threaten’.